Case Study — AtlasIQ

Executive Reporting Platform

Replacing a 40-Page Weekly PowerPoint With a Live Intelligence Dashboard

From static PDF decks to real-time C-suite intelligence in 10 weeks

Industry

Enterprise / SaaS

Timeline

10 weeks

Team

4 engineers

Tech

React + WebSocket + GPT-4

The Challenge

A 2,000-person SaaS company was producing a 40-page static PowerPoint deck every Monday morning — a snapshot of the previous week with no drill-down, no filtering, and no ability to ask follow-up questions. Three analysts spent their entire Monday building it instead of doing analysis.

Our Approach

How We Solved It

KPI Taxonomy & Ownership Mapping

Ran a 2-day workshop with the C-suite to define the 22 KPIs that actually drive decisions, who owns each metric, and what 'good' looks like — before writing a line of code.

Live Data Integration

Connected Salesforce, HubSpot, Stripe, Looker, and their internal PostgreSQL data warehouse via a unified API layer that refreshes every 15 minutes.

Narrative AI Layer

GPT-4 generates a 3-paragraph executive summary every Monday morning highlighting the 5 biggest movements, their likely causes, and items requiring CEO attention.

Mobile-First Dashboard Design

Built the dashboard mobile-first since the CEO reviews metrics from an iPad before the Monday leadership meeting. Swipe gestures and large tap targets reduced interaction friction by 60%.

Engineering Process

How We Built It

Push vs Pull Architecture

Used WebSocket push for live KPI tiles so executives see metric changes in real time without refreshing — critical for active leadership meetings.

Caching Strategy

Implemented a three-tier cache (L1: Redis for live tiles, L2: PostgreSQL materialized views for period aggregations, L3: S3 snapshots for historical comparisons).

Permission Scoping

Each executive sees a version of the dashboard scoped to their function — CFO sees full financials, CMO sees marketing + revenue — with one shared CEO view.

Architecture Decisions

Key Technical Choices

Purpose-Built Over Embedded BI

Chose a custom React build over embedding Metabase or Tableau because the CEO's specific workflow required interactions and narrative features no off-the-shelf BI tool supports.

GPT Summary as Supplement, Not Replacement

AI narrative is one section, not the whole dashboard. We surfaced it as 'This Week's Highlights' so executives still see raw numbers and can override the AI interpretation.

Snapshot Archive for Trend Comparison

Scheduled daily snapshots of all 22 KPIs to S3 so executives can compare current state to any historical Monday with a single click.

Results

What We Delivered

100%

Elimination of Manual Reports

Real-time

Dashboard Updates (was weekly)

85%

Reduction in Analyst Prep Time

3 FTEs

Analyst Capacity Freed

Solution Blueprint

How It All Fits Together

Data Integration Layer

Salesforce CRM connector
Stripe billing connector
Internal data warehouse sync

Intelligence Layer

GPT-4 narrative generation
Anomaly detection
Trend signal extraction

Presentation Layer

Mobile-first React dashboard
WebSocket live tiles
Drill-down explorer

Lessons Learned

What We Improved

The KPI Workshop Is the Product

The hardest part wasn't engineering — it was getting 6 C-suite executives to agree on 22 KPIs. That alignment work is what made the dashboard actually used.

AI Narrative Needs Human Framing

The first GPT summaries were too verbose. We iterated to a strict 3-paragraph format: What happened, Why it happened, What to do — and adoption doubled.

Mobile Testing Is Non-Negotiable

We discovered our CEO actually reviewed metrics during 6am walks. Pixel-perfect desktop views that are unreadable on mobile would have killed adoption immediately.